Subspace Identification
-----------------------
This diskette goes together with the book:
Subspace Identification for Linear Systems
Theory - Implementation - Applications
By Peter Van Overschee and Bart De Moor
Kluwer Academic Publishers, 1996
It contains all algorithms described in the book.
Installation
------------
All algorithms in the book have been implemented in separate M-files.
The research however culminated in the function subid.m which
implements deterministic, stochastic and combined subspace
identification. This function is an excellent place to start with.
After you have copied the files from the diskette to your
computer (for instance in the directory c:\subspace), you should
include the directory ;subfun' in your path:
> path(path,'c:\subspace\subfun');
Now change your directory to the 'examples' directory and
run the demo file 'sta_demo.m':
> cd c:\subspace\examples
> sta_demo
You will get a clear impression of what subspace identification
algorithms can do.
Now you are ready to rerun some of the applications. To run for
instance the application of the flexible robot arm, do:
> cd c:\subspace\applic
> appl5
This will guide you through the demo and reproduce the results of the
book. Now that you are at it, you could run some other applications
in this directory.
\nin You are now ready to try out your own problem. You could just
use one of the subspace identification functions (for instance
'subid.m'. Alternatively, you could also copy and adapt one of the
application M-files.
Overview:
---------
For practical applications, only the function 'subid' is of
importance. The other subspace identification functions that are
included, are more of an academic nature:
Deterministic Identification: det_stat, det_alt, intersec, project / subid
Stochastic Identification: sto_stat, sto_alt, sto_pos / subid
Combined Identification: com_stat, com_alt / subid
For more information on the stochastic identification, see the file
examples/sto_demo.m.
Notes:
------
It should be noted that there is still room to improve the speed and
the memory usage of the algorithms, using for instance displacement
structure algorithms for the RQ decomposition and Lanczos algorithms
for the SVD. The algorithms that have been implemented on this
diskette only use basic MATLAB commands (and no fancy optimized
C-code). Feel free to improve, and let me know what you think:
peter.vanoverschee@esat.kuleuven.ac.be